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#include <linux/swapops.h>
21#include <linux/sysctl.h>
22#include <linux/ksm.h>
23#include <linux/mman.h>
24
25#include <asm/pgtable.h>
26#include <asm/pgalloc.h>
27#include <asm/tlb.h>
28#include <asm/tlbflush.h>
29#include <asm/mmu_context.h>
30
31#define ALLOC_ORDER	2
32#define FRAG_MASK	0x03
33
34unsigned long *crst_table_alloc(struct mm_struct *mm)
35{
36	struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
37
38	if (!page)
39		return NULL;
40	return (unsigned long *) page_to_phys(page);
41}
42
43void crst_table_free(struct mm_struct *mm, unsigned long *table)
44{
45	free_pages((unsigned long) table, ALLOC_ORDER);
46}
47
48static void __crst_table_upgrade(void *arg)
49{
50	struct mm_struct *mm = arg;
51
52	if (current->active_mm == mm) {
53		clear_user_asce();
54		set_user_asce(mm);
55	}
56	__tlb_flush_local();
57}
58
59int crst_table_upgrade(struct mm_struct *mm, unsigned long limit)
60{
61	unsigned long *table, *pgd;
62	unsigned long entry;
63	int flush;
64
65	BUG_ON(limit > (1UL << 53));
66	flush = 0;
67repeat:
68	table = crst_table_alloc(mm);
69	if (!table)
70		return -ENOMEM;
71	spin_lock_bh(&mm->page_table_lock);
72	if (mm->context.asce_limit < limit) {
73		pgd = (unsigned long *) mm->pgd;
74		if (mm->context.asce_limit <= (1UL << 31)) {
75			entry = _REGION3_ENTRY_EMPTY;
76			mm->context.asce_limit = 1UL << 42;
77			mm->context.asce_bits = _ASCE_TABLE_LENGTH |
78						_ASCE_USER_BITS |
79						_ASCE_TYPE_REGION3;
80		} else {
81			entry = _REGION2_ENTRY_EMPTY;
82			mm->context.asce_limit = 1UL << 53;
83			mm->context.asce_bits = _ASCE_TABLE_LENGTH |
84						_ASCE_USER_BITS |
85						_ASCE_TYPE_REGION2;
86		}
87		crst_table_init(table, entry);
88		pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd);
89		mm->pgd = (pgd_t *) table;
90		mm->task_size = mm->context.asce_limit;
91		table = NULL;
92		flush = 1;
93	}
94	spin_unlock_bh(&mm->page_table_lock);
95	if (table)
96		crst_table_free(mm, table);
97	if (mm->context.asce_limit < limit)
98		goto repeat;
99	if (flush)
100		on_each_cpu(__crst_table_upgrade, mm, 0);
101	return 0;
102}
103
104void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
105{
106	pgd_t *pgd;
107
108	if (current->active_mm == mm) {
109		clear_user_asce();
110		__tlb_flush_mm(mm);
111	}
112	while (mm->context.asce_limit > limit) {
113		pgd = mm->pgd;
114		switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) {
115		case _REGION_ENTRY_TYPE_R2:
116			mm->context.asce_limit = 1UL << 42;
117			mm->context.asce_bits = _ASCE_TABLE_LENGTH |
118						_ASCE_USER_BITS |
119						_ASCE_TYPE_REGION3;
120			break;
121		case _REGION_ENTRY_TYPE_R3:
122			mm->context.asce_limit = 1UL << 31;
123			mm->context.asce_bits = _ASCE_TABLE_LENGTH |
124						_ASCE_USER_BITS |
125						_ASCE_TYPE_SEGMENT;
126			break;
127		default:
128			BUG();
129		}
130		mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
131		mm->task_size = mm->context.asce_limit;
132		crst_table_free(mm, (unsigned long *) pgd);
133	}
134	if (current->active_mm == mm)
135		set_user_asce(mm);
136}
137
138#ifdef CONFIG_PGSTE
139
140/**
141 * gmap_alloc - allocate a guest address space
142 * @mm: pointer to the parent mm_struct
143 * @limit: maximum size of the gmap address space
144 *
145 * Returns a guest address space structure.
146 */
147struct gmap *gmap_alloc(struct mm_struct *mm, unsigned long limit)
148{
149	struct gmap *gmap;
150	struct page *page;
151	unsigned long *table;
152	unsigned long etype, atype;
153
154	if (limit < (1UL << 31)) {
155		limit = (1UL << 31) - 1;
156		atype = _ASCE_TYPE_SEGMENT;
157		etype = _SEGMENT_ENTRY_EMPTY;
158	} else if (limit < (1UL << 42)) {
159		limit = (1UL << 42) - 1;
160		atype = _ASCE_TYPE_REGION3;
161		etype = _REGION3_ENTRY_EMPTY;
162	} else if (limit < (1UL << 53)) {
163		limit = (1UL << 53) - 1;
164		atype = _ASCE_TYPE_REGION2;
165		etype = _REGION2_ENTRY_EMPTY;
166	} else {
167		limit = -1UL;
168		atype = _ASCE_TYPE_REGION1;
169		etype = _REGION1_ENTRY_EMPTY;
170	}
171	gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
172	if (!gmap)
173		goto out;
174	INIT_LIST_HEAD(&gmap->crst_list);
175	INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL);
176	INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC);
177	spin_lock_init(&gmap->guest_table_lock);
178	gmap->mm = mm;
179	page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
180	if (!page)
181		goto out_free;
182	page->index = 0;
183	list_add(&page->lru, &gmap->crst_list);
184	table = (unsigned long *) page_to_phys(page);
185	crst_table_init(table, etype);
186	gmap->table = table;
187	gmap->asce = atype | _ASCE_TABLE_LENGTH |
188		_ASCE_USER_BITS | __pa(table);
189	gmap->asce_end = limit;
190	down_write(&mm->mmap_sem);
191	list_add(&gmap->list, &mm->context.gmap_list);
192	up_write(&mm->mmap_sem);
193	return gmap;
194
195out_free:
196	kfree(gmap);
197out:
198	return NULL;
199}
200EXPORT_SYMBOL_GPL(gmap_alloc);
201
202static void gmap_flush_tlb(struct gmap *gmap)
203{
204	if (MACHINE_HAS_IDTE)
205		__tlb_flush_asce(gmap->mm, gmap->asce);
206	else
207		__tlb_flush_global();
208}
209
210static void gmap_radix_tree_free(struct radix_tree_root *root)
211{
212	struct radix_tree_iter iter;
213	unsigned long indices[16];
214	unsigned long index;
215	void **slot;
216	int i, nr;
217
218	/* A radix tree is freed by deleting all of its entries */
219	index = 0;
220	do {
221		nr = 0;
222		radix_tree_for_each_slot(slot, root, &iter, index) {
223			indices[nr] = iter.index;
224			if (++nr == 16)
225				break;
226		}
227		for (i = 0; i < nr; i++) {
228			index = indices[i];
229			radix_tree_delete(root, index);
230		}
231	} while (nr > 0);
232}
233
234/**
235 * gmap_free - free a guest address space
236 * @gmap: pointer to the guest address space structure
237 */
238void gmap_free(struct gmap *gmap)
239{
240	struct page *page, *next;
241
242	/* Flush tlb. */
243	if (MACHINE_HAS_IDTE)
244		__tlb_flush_asce(gmap->mm, gmap->asce);
245	else
246		__tlb_flush_global();
247
248	/* Free all segment & region tables. */
249	list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
250		__free_pages(page, ALLOC_ORDER);
251	gmap_radix_tree_free(&gmap->guest_to_host);
252	gmap_radix_tree_free(&gmap->host_to_guest);
253	down_write(&gmap->mm->mmap_sem);
254	list_del(&gmap->list);
255	up_write(&gmap->mm->mmap_sem);
256	kfree(gmap);
257}
258EXPORT_SYMBOL_GPL(gmap_free);
259
260/**
261 * gmap_enable - switch primary space to the guest address space
262 * @gmap: pointer to the guest address space structure
263 */
264void gmap_enable(struct gmap *gmap)
265{
266	S390_lowcore.gmap = (unsigned long) gmap;
267}
268EXPORT_SYMBOL_GPL(gmap_enable);
269
270/**
271 * gmap_disable - switch back to the standard primary address space
272 * @gmap: pointer to the guest address space structure
273 */
274void gmap_disable(struct gmap *gmap)
275{
276	S390_lowcore.gmap = 0UL;
277}
278EXPORT_SYMBOL_GPL(gmap_disable);
279
280/*
281 * gmap_alloc_table is assumed to be called with mmap_sem held
282 */
283static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
284			    unsigned long init, unsigned long gaddr)
285{
286	struct page *page;
287	unsigned long *new;
288
289	/* since we dont free the gmap table until gmap_free we can unlock */
290	page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
291	if (!page)
292		return -ENOMEM;
293	new = (unsigned long *) page_to_phys(page);
294	crst_table_init(new, init);
295	spin_lock(&gmap->mm->page_table_lock);
296	if (*table & _REGION_ENTRY_INVALID) {
297		list_add(&page->lru, &gmap->crst_list);
298		*table = (unsigned long) new | _REGION_ENTRY_LENGTH |
299			(*table & _REGION_ENTRY_TYPE_MASK);
300		page->index = gaddr;
301		page = NULL;
302	}
303	spin_unlock(&gmap->mm->page_table_lock);
304	if (page)
305		__free_pages(page, ALLOC_ORDER);
306	return 0;
307}
308
309/**
310 * __gmap_segment_gaddr - find virtual address from segment pointer
311 * @entry: pointer to a segment table entry in the guest address space
312 *
313 * Returns the virtual address in the guest address space for the segment
314 */
315static unsigned long __gmap_segment_gaddr(unsigned long *entry)
316{
317	struct page *page;
318	unsigned long offset, mask;
319
320	offset = (unsigned long) entry / sizeof(unsigned long);
321	offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
322	mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
323	page = virt_to_page((void *)((unsigned long) entry & mask));
324	return page->index + offset;
325}
326
327/**
328 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
329 * @gmap: pointer to the guest address space structure
330 * @vmaddr: address in the host process address space
331 *
332 * Returns 1 if a TLB flush is required
333 */
334static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
335{
336	unsigned long *entry;
337	int flush = 0;
338
339	spin_lock(&gmap->guest_table_lock);
340	entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
341	if (entry) {
342		flush = (*entry != _SEGMENT_ENTRY_INVALID);
343		*entry = _SEGMENT_ENTRY_INVALID;
344	}
345	spin_unlock(&gmap->guest_table_lock);
346	return flush;
347}
348
349/**
350 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
351 * @gmap: pointer to the guest address space structure
352 * @gaddr: address in the guest address space
353 *
354 * Returns 1 if a TLB flush is required
355 */
356static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
357{
358	unsigned long vmaddr;
359
360	vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
361						   gaddr >> PMD_SHIFT);
362	return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
363}
364
365/**
366 * gmap_unmap_segment - unmap segment from the guest address space
367 * @gmap: pointer to the guest address space structure
368 * @to: address in the guest address space
369 * @len: length of the memory area to unmap
370 *
371 * Returns 0 if the unmap succeeded, -EINVAL if not.
372 */
373int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
374{
375	unsigned long off;
376	int flush;
377
378	if ((to | len) & (PMD_SIZE - 1))
379		return -EINVAL;
380	if (len == 0 || to + len < to)
381		return -EINVAL;
382
383	flush = 0;
384	down_write(&gmap->mm->mmap_sem);
385	for (off = 0; off < len; off += PMD_SIZE)
386		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
387	up_write(&gmap->mm->mmap_sem);
388	if (flush)
389		gmap_flush_tlb(gmap);
390	return 0;
391}
392EXPORT_SYMBOL_GPL(gmap_unmap_segment);
393
394/**
395 * gmap_mmap_segment - map a segment to the guest address space
396 * @gmap: pointer to the guest address space structure
397 * @from: source address in the parent address space
398 * @to: target address in the guest address space
399 * @len: length of the memory area to map
400 *
401 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
402 */
403int gmap_map_segment(struct gmap *gmap, unsigned long from,
404		     unsigned long to, unsigned long len)
405{
406	unsigned long off;
407	int flush;
408
409	if ((from | to | len) & (PMD_SIZE - 1))
410		return -EINVAL;
411	if (len == 0 || from + len < from || to + len < to ||
412	    from + len > TASK_MAX_SIZE || to + len > gmap->asce_end)
413		return -EINVAL;
414
415	flush = 0;
416	down_write(&gmap->mm->mmap_sem);
417	for (off = 0; off < len; off += PMD_SIZE) {
418		/* Remove old translation */
419		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
420		/* Store new translation */
421		if (radix_tree_insert(&gmap->guest_to_host,
422				      (to + off) >> PMD_SHIFT,
423				      (void *) from + off))
424			break;
425	}
426	up_write(&gmap->mm->mmap_sem);
427	if (flush)
428		gmap_flush_tlb(gmap);
429	if (off >= len)
430		return 0;
431	gmap_unmap_segment(gmap, to, len);
432	return -ENOMEM;
433}
434EXPORT_SYMBOL_GPL(gmap_map_segment);
435
436/**
437 * __gmap_translate - translate a guest address to a user space address
438 * @gmap: pointer to guest mapping meta data structure
439 * @gaddr: guest address
440 *
441 * Returns user space address which corresponds to the guest address or
442 * -EFAULT if no such mapping exists.
443 * This function does not establish potentially missing page table entries.
444 * The mmap_sem of the mm that belongs to the address space must be held
445 * when this function gets called.
446 */
447unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
448{
449	unsigned long vmaddr;
450
451	vmaddr = (unsigned long)
452		radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
453	return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
454}
455EXPORT_SYMBOL_GPL(__gmap_translate);
456
457/**
458 * gmap_translate - translate a guest address to a user space address
459 * @gmap: pointer to guest mapping meta data structure
460 * @gaddr: guest address
461 *
462 * Returns user space address which corresponds to the guest address or
463 * -EFAULT if no such mapping exists.
464 * This function does not establish potentially missing page table entries.
465 */
466unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
467{
468	unsigned long rc;
469
470	down_read(&gmap->mm->mmap_sem);
471	rc = __gmap_translate(gmap, gaddr);
472	up_read(&gmap->mm->mmap_sem);
473	return rc;
474}
475EXPORT_SYMBOL_GPL(gmap_translate);
476
477/**
478 * gmap_unlink - disconnect a page table from the gmap shadow tables
479 * @gmap: pointer to guest mapping meta data structure
480 * @table: pointer to the host page table
481 * @vmaddr: vm address associated with the host page table
482 */
483static void gmap_unlink(struct mm_struct *mm, unsigned long *table,
484			unsigned long vmaddr)
485{
486	struct gmap *gmap;
487	int flush;
488
489	list_for_each_entry(gmap, &mm->context.gmap_list, list) {
490		flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
491		if (flush)
492			gmap_flush_tlb(gmap);
493	}
494}
495
496/**
497 * gmap_link - set up shadow page tables to connect a host to a guest address
498 * @gmap: pointer to guest mapping meta data structure
499 * @gaddr: guest address
500 * @vmaddr: vm address
501 *
502 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
503 * if the vm address is already mapped to a different guest segment.
504 * The mmap_sem of the mm that belongs to the address space must be held
505 * when this function gets called.
506 */
507int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
508{
509	struct mm_struct *mm;
510	unsigned long *table;
511	spinlock_t *ptl;
512	pgd_t *pgd;
513	pud_t *pud;
514	pmd_t *pmd;
515	int rc;
516
517	/* Create higher level tables in the gmap page table */
518	table = gmap->table;
519	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
520		table += (gaddr >> 53) & 0x7ff;
521		if ((*table & _REGION_ENTRY_INVALID) &&
522		    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
523				     gaddr & 0xffe0000000000000UL))
524			return -ENOMEM;
525		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
526	}
527	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
528		table += (gaddr >> 42) & 0x7ff;
529		if ((*table & _REGION_ENTRY_INVALID) &&
530		    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
531				     gaddr & 0xfffffc0000000000UL))
532			return -ENOMEM;
533		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
534	}
535	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
536		table += (gaddr >> 31) & 0x7ff;
537		if ((*table & _REGION_ENTRY_INVALID) &&
538		    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
539				     gaddr & 0xffffffff80000000UL))
540			return -ENOMEM;
541		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
542	}
543	table += (gaddr >> 20) & 0x7ff;
544	/* Walk the parent mm page table */
545	mm = gmap->mm;
546	pgd = pgd_offset(mm, vmaddr);
547	VM_BUG_ON(pgd_none(*pgd));
548	pud = pud_offset(pgd, vmaddr);
549	VM_BUG_ON(pud_none(*pud));
550	pmd = pmd_offset(pud, vmaddr);
551	VM_BUG_ON(pmd_none(*pmd));
552	/* large pmds cannot yet be handled */
553	if (pmd_large(*pmd))
554		return -EFAULT;
555	/* Link gmap segment table entry location to page table. */
556	rc = radix_tree_preload(GFP_KERNEL);
557	if (rc)
558		return rc;
559	ptl = pmd_lock(mm, pmd);
560	spin_lock(&gmap->guest_table_lock);
561	if (*table == _SEGMENT_ENTRY_INVALID) {
562		rc = radix_tree_insert(&gmap->host_to_guest,
563				       vmaddr >> PMD_SHIFT, table);
564		if (!rc)
565			*table = pmd_val(*pmd);
566	} else
567		rc = 0;
568	spin_unlock(&gmap->guest_table_lock);
569	spin_unlock(ptl);
570	radix_tree_preload_end();
571	return rc;
572}
573
574/**
575 * gmap_fault - resolve a fault on a guest address
576 * @gmap: pointer to guest mapping meta data structure
577 * @gaddr: guest address
578 * @fault_flags: flags to pass down to handle_mm_fault()
579 *
580 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
581 * if the vm address is already mapped to a different guest segment.
582 */
583int gmap_fault(struct gmap *gmap, unsigned long gaddr,
584	       unsigned int fault_flags)
585{
586	unsigned long vmaddr;
587	int rc;
588
589	down_read(&gmap->mm->mmap_sem);
590	vmaddr = __gmap_translate(gmap, gaddr);
591	if (IS_ERR_VALUE(vmaddr)) {
592		rc = vmaddr;
593		goto out_up;
594	}
595	if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags)) {
596		rc = -EFAULT;
597		goto out_up;
598	}
599	rc = __gmap_link(gmap, gaddr, vmaddr);
600out_up:
601	up_read(&gmap->mm->mmap_sem);
602	return rc;
603}
604EXPORT_SYMBOL_GPL(gmap_fault);
605
606static void gmap_zap_swap_entry(swp_entry_t entry, struct mm_struct *mm)
607{
608	if (!non_swap_entry(entry))
609		dec_mm_counter(mm, MM_SWAPENTS);
610	else if (is_migration_entry(entry)) {
611		struct page *page = migration_entry_to_page(entry);
612
613		if (PageAnon(page))
614			dec_mm_counter(mm, MM_ANONPAGES);
615		else
616			dec_mm_counter(mm, MM_FILEPAGES);
617	}
618	free_swap_and_cache(entry);
619}
620
621/*
622 * this function is assumed to be called with mmap_sem held
623 */
624void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
625{
626	unsigned long vmaddr, ptev, pgstev;
627	pte_t *ptep, pte;
628	spinlock_t *ptl;
629	pgste_t pgste;
630
631	/* Find the vm address for the guest address */
632	vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
633						   gaddr >> PMD_SHIFT);
634	if (!vmaddr)
635		return;
636	vmaddr |= gaddr & ~PMD_MASK;
637	/* Get pointer to the page table entry */
638	ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
639	if (unlikely(!ptep))
640		return;
641	pte = *ptep;
642	if (!pte_swap(pte))
643		goto out_pte;
644	/* Zap unused and logically-zero pages */
645	pgste = pgste_get_lock(ptep);
646	pgstev = pgste_val(pgste);
647	ptev = pte_val(pte);
648	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
649	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) {
650		gmap_zap_swap_entry(pte_to_swp_entry(pte), gmap->mm);
651		pte_clear(gmap->mm, vmaddr, ptep);
652	}
653	pgste_set_unlock(ptep, pgste);
654out_pte:
655	pte_unmap_unlock(ptep, ptl);
656}
657EXPORT_SYMBOL_GPL(__gmap_zap);
658
659void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
660{
661	unsigned long gaddr, vmaddr, size;
662	struct vm_area_struct *vma;
663
664	down_read(&gmap->mm->mmap_sem);
665	for (gaddr = from; gaddr < to;
666	     gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
667		/* Find the vm address for the guest address */
668		vmaddr = (unsigned long)
669			radix_tree_lookup(&gmap->guest_to_host,
670					  gaddr >> PMD_SHIFT);
671		if (!vmaddr)
672			continue;
673		vmaddr |= gaddr & ~PMD_MASK;
674		/* Find vma in the parent mm */
675		vma = find_vma(gmap->mm, vmaddr);
676		size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
677		zap_page_range(vma, vmaddr, size, NULL);
678	}
679	up_read(&gmap->mm->mmap_sem);
680}
681EXPORT_SYMBOL_GPL(gmap_discard);
682
683static LIST_HEAD(gmap_notifier_list);
684static DEFINE_SPINLOCK(gmap_notifier_lock);
685
686/**
687 * gmap_register_ipte_notifier - register a pte invalidation callback
688 * @nb: pointer to the gmap notifier block
689 */
690void gmap_register_ipte_notifier(struct gmap_notifier *nb)
691{
692	spin_lock(&gmap_notifier_lock);
693	list_add(&nb->list, &gmap_notifier_list);
694	spin_unlock(&gmap_notifier_lock);
695}
696EXPORT_SYMBOL_GPL(gmap_register_ipte_notifier);
697
698/**
699 * gmap_unregister_ipte_notifier - remove a pte invalidation callback
700 * @nb: pointer to the gmap notifier block
701 */
702void gmap_unregister_ipte_notifier(struct gmap_notifier *nb)
703{
704	spin_lock(&gmap_notifier_lock);
705	list_del_init(&nb->list);
706	spin_unlock(&gmap_notifier_lock);
707}
708EXPORT_SYMBOL_GPL(gmap_unregister_ipte_notifier);
709
710/**
711 * gmap_ipte_notify - mark a range of ptes for invalidation notification
712 * @gmap: pointer to guest mapping meta data structure
713 * @gaddr: virtual address in the guest address space
714 * @len: size of area
715 *
716 * Returns 0 if for each page in the given range a gmap mapping exists and
717 * the invalidation notification could be set. If the gmap mapping is missing
718 * for one or more pages -EFAULT is returned. If no memory could be allocated
719 * -ENOMEM is returned. This function establishes missing page table entries.
720 */
721int gmap_ipte_notify(struct gmap *gmap, unsigned long gaddr, unsigned long len)
722{
723	unsigned long addr;
724	spinlock_t *ptl;
725	pte_t *ptep, entry;
726	pgste_t pgste;
727	int rc = 0;
728
729	if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK))
730		return -EINVAL;
731	down_read(&gmap->mm->mmap_sem);
732	while (len) {
733		/* Convert gmap address and connect the page tables */
734		addr = __gmap_translate(gmap, gaddr);
735		if (IS_ERR_VALUE(addr)) {
736			rc = addr;
737			break;
738		}
739		/* Get the page mapped */
740		if (fixup_user_fault(current, gmap->mm, addr, FAULT_FLAG_WRITE)) {
741			rc = -EFAULT;
742			break;
743		}
744		rc = __gmap_link(gmap, gaddr, addr);
745		if (rc)
746			break;
747		/* Walk the process page table, lock and get pte pointer */
748		ptep = get_locked_pte(gmap->mm, addr, &ptl);
749		VM_BUG_ON(!ptep);
750		/* Set notification bit in the pgste of the pte */
751		entry = *ptep;
752		if ((pte_val(entry) & (_PAGE_INVALID | _PAGE_PROTECT)) == 0) {
753			pgste = pgste_get_lock(ptep);
754			pgste_val(pgste) |= PGSTE_IN_BIT;
755			pgste_set_unlock(ptep, pgste);
756			gaddr += PAGE_SIZE;
757			len -= PAGE_SIZE;
758		}
759		pte_unmap_unlock(ptep, ptl);
760	}
761	up_read(&gmap->mm->mmap_sem);
762	return rc;
763}
764EXPORT_SYMBOL_GPL(gmap_ipte_notify);
765
766/**
767 * gmap_do_ipte_notify - call all invalidation callbacks for a specific pte.
768 * @mm: pointer to the process mm_struct
769 * @addr: virtual address in the process address space
770 * @pte: pointer to the page table entry
771 *
772 * This function is assumed to be called with the page table lock held
773 * for the pte to notify.
774 */
775void gmap_do_ipte_notify(struct mm_struct *mm, unsigned long vmaddr, pte_t *pte)
776{
777	unsigned long offset, gaddr;
778	unsigned long *table;
779	struct gmap_notifier *nb;
780	struct gmap *gmap;
781
782	offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
783	offset = offset * (4096 / sizeof(pte_t));
784	spin_lock(&gmap_notifier_lock);
785	list_for_each_entry(gmap, &mm->context.gmap_list, list) {
786		table = radix_tree_lookup(&gmap->host_to_guest,
787					  vmaddr >> PMD_SHIFT);
788		if (!table)
789			continue;
790		gaddr = __gmap_segment_gaddr(table) + offset;
791		list_for_each_entry(nb, &gmap_notifier_list, list)
792			nb->notifier_call(gmap, gaddr);
793	}
794	spin_unlock(&gmap_notifier_lock);
795}
796EXPORT_SYMBOL_GPL(gmap_do_ipte_notify);
797
798static inline int page_table_with_pgste(struct page *page)
799{
800	return atomic_read(&page->_mapcount) == 0;
801}
802
803static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm)
804{
805	struct page *page;
806	unsigned long *table;
807
808	page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
809	if (!page)
810		return NULL;
811	if (!pgtable_page_ctor(page)) {
812		__free_page(page);
813		return NULL;
814	}
815	atomic_set(&page->_mapcount, 0);
816	table = (unsigned long *) page_to_phys(page);
817	clear_table(table, _PAGE_INVALID, PAGE_SIZE/2);
818	clear_table(table + PTRS_PER_PTE, 0, PAGE_SIZE/2);
819	return table;
820}
821
822static inline void page_table_free_pgste(unsigned long *table)
823{
824	struct page *page;
825
826	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
827	pgtable_page_dtor(page);
828	atomic_set(&page->_mapcount, -1);
829	__free_page(page);
830}
831
832int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
833			  unsigned long key, bool nq)
834{
835	spinlock_t *ptl;
836	pgste_t old, new;
837	pte_t *ptep;
838
839	down_read(&mm->mmap_sem);
840retry:
841	ptep = get_locked_pte(mm, addr, &ptl);
842	if (unlikely(!ptep)) {
843		up_read(&mm->mmap_sem);
844		return -EFAULT;
845	}
846	if (!(pte_val(*ptep) & _PAGE_INVALID) &&
847	     (pte_val(*ptep) & _PAGE_PROTECT)) {
848		pte_unmap_unlock(ptep, ptl);
849		if (fixup_user_fault(current, mm, addr, FAULT_FLAG_WRITE)) {
850			up_read(&mm->mmap_sem);
851			return -EFAULT;
852		}
853		goto retry;
854	}
855
856	new = old = pgste_get_lock(ptep);
857	pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
858			    PGSTE_ACC_BITS | PGSTE_FP_BIT);
859	pgste_val(new) |= (key & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
860	pgste_val(new) |= (key & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
861	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
862		unsigned long address, bits, skey;
863
864		address = pte_val(*ptep) & PAGE_MASK;
865		skey = (unsigned long) page_get_storage_key(address);
866		bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
867		skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
868		/* Set storage key ACC and FP */
869		page_set_storage_key(address, skey, !nq);
870		/* Merge host changed & referenced into pgste  */
871		pgste_val(new) |= bits << 52;
872	}
873	/* changing the guest storage key is considered a change of the page */
874	if ((pgste_val(new) ^ pgste_val(old)) &
875	    (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
876		pgste_val(new) |= PGSTE_UC_BIT;
877
878	pgste_set_unlock(ptep, new);
879	pte_unmap_unlock(ptep, ptl);
880	up_read(&mm->mmap_sem);
881	return 0;
882}
883EXPORT_SYMBOL(set_guest_storage_key);
884
885unsigned long get_guest_storage_key(struct mm_struct *mm, unsigned long addr)
886{
887	spinlock_t *ptl;
888	pgste_t pgste;
889	pte_t *ptep;
890	uint64_t physaddr;
891	unsigned long key = 0;
892
893	down_read(&mm->mmap_sem);
894	ptep = get_locked_pte(mm, addr, &ptl);
895	if (unlikely(!ptep)) {
896		up_read(&mm->mmap_sem);
897		return -EFAULT;
898	}
899	pgste = pgste_get_lock(ptep);
900
901	if (pte_val(*ptep) & _PAGE_INVALID) {
902		key |= (pgste_val(pgste) & PGSTE_ACC_BITS) >> 56;
903		key |= (pgste_val(pgste) & PGSTE_FP_BIT) >> 56;
904		key |= (pgste_val(pgste) & PGSTE_GR_BIT) >> 48;
905		key |= (pgste_val(pgste) & PGSTE_GC_BIT) >> 48;
906	} else {
907		physaddr = pte_val(*ptep) & PAGE_MASK;
908		key = page_get_storage_key(physaddr);
909
910		/* Reflect guest's logical view, not physical */
911		if (pgste_val(pgste) & PGSTE_GR_BIT)
912			key |= _PAGE_REFERENCED;
913		if (pgste_val(pgste) & PGSTE_GC_BIT)
914			key |= _PAGE_CHANGED;
915	}
916
917	pgste_set_unlock(ptep, pgste);
918	pte_unmap_unlock(ptep, ptl);
919	up_read(&mm->mmap_sem);
920	return key;
921}
922EXPORT_SYMBOL(get_guest_storage_key);
923
924static int page_table_allocate_pgste_min = 0;
925static int page_table_allocate_pgste_max = 1;
926int page_table_allocate_pgste = 0;
927EXPORT_SYMBOL(page_table_allocate_pgste);
928
929static struct ctl_table page_table_sysctl[] = {
930	{
931		.procname	= "allocate_pgste",
932		.data		= &page_table_allocate_pgste,
933		.maxlen		= sizeof(int),
934		.mode		= S_IRUGO | S_IWUSR,
935		.proc_handler	= proc_dointvec,
936		.extra1		= &page_table_allocate_pgste_min,
937		.extra2		= &page_table_allocate_pgste_max,
938	},
939	{ }
940};
941
942static struct ctl_table page_table_sysctl_dir[] = {
943	{
944		.procname	= "vm",
945		.maxlen		= 0,
946		.mode		= 0555,
947		.child		= page_table_sysctl,
948	},
949	{ }
950};
951
952static int __init page_table_register_sysctl(void)
953{
954	return register_sysctl_table(page_table_sysctl_dir) ? 0 : -ENOMEM;
955}
956__initcall(page_table_register_sysctl);
957
958#else /* CONFIG_PGSTE */
959
960static inline int page_table_with_pgste(struct page *page)
961{
962	return 0;
963}
964
965static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm)
966{
967	return NULL;
968}
969
970static inline void page_table_free_pgste(unsigned long *table)
971{
972}
973
974static inline void gmap_unlink(struct mm_struct *mm, unsigned long *table,
975			unsigned long vmaddr)
976{
977}
978
979#endif /* CONFIG_PGSTE */
980
981static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
982{
983	unsigned int old, new;
984
985	do {
986		old = atomic_read(v);
987		new = old ^ bits;
988	} while (atomic_cmpxchg(v, old, new) != old);
989	return new;
990}
991
992/*
993 * page table entry allocation/free routines.
994 */
995unsigned long *page_table_alloc(struct mm_struct *mm)
996{
997	unsigned long *uninitialized_var(table);
998	struct page *uninitialized_var(page);
999	unsigned int mask, bit;
1000
1001	if (mm_alloc_pgste(mm))
1002		return page_table_alloc_pgste(mm);
1003	/* Allocate fragments of a 4K page as 1K/2K page table */
1004	spin_lock_bh(&mm->context.list_lock);
1005	mask = FRAG_MASK;
1006	if (!list_empty(&mm->context.pgtable_list)) {
1007		page = list_first_entry(&mm->context.pgtable_list,
1008					struct page, lru);
1009		table = (unsigned long *) page_to_phys(page);
1010		mask = atomic_read(&page->_mapcount);
1011		mask = mask | (mask >> 4);
1012	}
1013	if ((mask & FRAG_MASK) == FRAG_MASK) {
1014		spin_unlock_bh(&mm->context.list_lock);
1015		page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
1016		if (!page)
1017			return NULL;
1018		if (!pgtable_page_ctor(page)) {
1019			__free_page(page);
1020			return NULL;
1021		}
1022		atomic_set(&page->_mapcount, 1);
1023		table = (unsigned long *) page_to_phys(page);
1024		clear_table(table, _PAGE_INVALID, PAGE_SIZE);
1025		spin_lock_bh(&mm->context.list_lock);
1026		list_add(&page->lru, &mm->context.pgtable_list);
1027	} else {
1028		for (bit = 1; mask & bit; bit <<= 1)
1029			table += PTRS_PER_PTE;
1030		mask = atomic_xor_bits(&page->_mapcount, bit);
1031		if ((mask & FRAG_MASK) == FRAG_MASK)
1032			list_del(&page->lru);
1033	}
1034	spin_unlock_bh(&mm->context.list_lock);
1035	return table;
1036}
1037
1038void page_table_free(struct mm_struct *mm, unsigned long *table)
1039{
1040	struct page *page;
1041	unsigned int bit, mask;
1042
1043	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
1044	if (page_table_with_pgste(page))
1045		return page_table_free_pgste(table);
1046	/* Free 1K/2K page table fragment of a 4K page */
1047	bit = 1 << ((__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t)));
1048	spin_lock_bh(&mm->context.list_lock);
1049	if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
1050		list_del(&page->lru);
1051	mask = atomic_xor_bits(&page->_mapcount, bit);
1052	if (mask & FRAG_MASK)
1053		list_add(&page->lru, &mm->context.pgtable_list);
1054	spin_unlock_bh(&mm->context.list_lock);
1055	if (mask == 0) {
1056		pgtable_page_dtor(page);
1057		atomic_set(&page->_mapcount, -1);
1058		__free_page(page);
1059	}
1060}
1061
1062static void __page_table_free_rcu(void *table, unsigned bit)
1063{
1064	struct page *page;
1065
1066	if (bit == FRAG_MASK)
1067		return page_table_free_pgste(table);
1068	/* Free 1K/2K page table fragment of a 4K page */
1069	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
1070	if (atomic_xor_bits(&page->_mapcount, bit) == 0) {
1071		pgtable_page_dtor(page);
1072		atomic_set(&page->_mapcount, -1);
1073		__free_page(page);
1074	}
1075}
1076
1077void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table,
1078			 unsigned long vmaddr)
1079{
1080	struct mm_struct *mm;
1081	struct page *page;
1082	unsigned int bit, mask;
1083
1084	mm = tlb->mm;
1085	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
1086	if (page_table_with_pgste(page)) {
1087		gmap_unlink(mm, table, vmaddr);
1088		table = (unsigned long *) (__pa(table) | FRAG_MASK);
1089		tlb_remove_table(tlb, table);
1090		return;
1091	}
1092	bit = 1 << ((__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t)));
1093	spin_lock_bh(&mm->context.list_lock);
1094	if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
1095		list_del(&page->lru);
1096	mask = atomic_xor_bits(&page->_mapcount, bit | (bit << 4));
1097	if (mask & FRAG_MASK)
1098		list_add_tail(&page->lru, &mm->context.pgtable_list);
1099	spin_unlock_bh(&mm->context.list_lock);
1100	table = (unsigned long *) (__pa(table) | (bit << 4));
1101	tlb_remove_table(tlb, table);
1102}
1103
1104static void __tlb_remove_table(void *_table)
1105{
1106	const unsigned long mask = (FRAG_MASK << 4) | FRAG_MASK;
1107	void *table = (void *)((unsigned long) _table & ~mask);
1108	unsigned type = (unsigned long) _table & mask;
1109
1110	if (type)
1111		__page_table_free_rcu(table, type);
1112	else
1113		free_pages((unsigned long) table, ALLOC_ORDER);
1114}
1115
1116static void tlb_remove_table_smp_sync(void *arg)
1117{
1118	/* Simply deliver the interrupt */
1119}
1120
1121static void tlb_remove_table_one(void *table)
1122{
1123	/*
1124	 * This isn't an RCU grace period and hence the page-tables cannot be
1125	 * assumed to be actually RCU-freed.
1126	 *
1127	 * It is however sufficient for software page-table walkers that rely
1128	 * on IRQ disabling. See the comment near struct mmu_table_batch.
1129	 */
1130	smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
1131	__tlb_remove_table(table);
1132}
1133
1134static void tlb_remove_table_rcu(struct rcu_head *head)
1135{
1136	struct mmu_table_batch *batch;
1137	int i;
1138
1139	batch = container_of(head, struct mmu_table_batch, rcu);
1140
1141	for (i = 0; i < batch->nr; i++)
1142		__tlb_remove_table(batch->tables[i]);
1143
1144	free_page((unsigned long)batch);
1145}
1146
1147void tlb_table_flush(struct mmu_gather *tlb)
1148{
1149	struct mmu_table_batch **batch = &tlb->batch;
1150
1151	if (*batch) {
1152		call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
1153		*batch = NULL;
1154	}
1155}
1156
1157void tlb_remove_table(struct mmu_gather *tlb, void *table)
1158{
1159	struct mmu_table_batch **batch = &tlb->batch;
1160
1161	tlb->mm->context.flush_mm = 1;
1162	if (*batch == NULL) {
1163		*batch = (struct mmu_table_batch *)
1164			__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
1165		if (*batch == NULL) {
1166			__tlb_flush_mm_lazy(tlb->mm);
1167			tlb_remove_table_one(table);
1168			return;
1169		}
1170		(*batch)->nr = 0;
1171	}
1172	(*batch)->tables[(*batch)->nr++] = table;
1173	if ((*batch)->nr == MAX_TABLE_BATCH)
1174		tlb_flush_mmu(tlb);
1175}
1176
1177#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1178static inline void thp_split_vma(struct vm_area_struct *vma)
1179{
1180	unsigned long addr;
1181
1182	for (addr = vma->vm_start; addr < vma->vm_end; addr += PAGE_SIZE)
1183		follow_page(vma, addr, FOLL_SPLIT);
1184}
1185
1186static inline void thp_split_mm(struct mm_struct *mm)
1187{
1188	struct vm_area_struct *vma;
1189
1190	for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
1191		thp_split_vma(vma);
1192		vma->vm_flags &= ~VM_HUGEPAGE;
1193		vma->vm_flags |= VM_NOHUGEPAGE;
1194	}
1195	mm->def_flags |= VM_NOHUGEPAGE;
1196}
1197#else
1198static inline void thp_split_mm(struct mm_struct *mm)
1199{
1200}
1201#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1202
1203/*
1204 * switch on pgstes for its userspace process (for kvm)
1205 */
1206int s390_enable_sie(void)
1207{
1208	struct mm_struct *mm = current->mm;
1209
1210	/* Do we have pgstes? if yes, we are done */
1211	if (mm_has_pgste(mm))
1212		return 0;
1213	/* Fail if the page tables are 2K */
1214	if (!mm_alloc_pgste(mm))
1215		return -EINVAL;
1216	down_write(&mm->mmap_sem);
1217	mm->context.has_pgste = 1;
1218	/* split thp mappings and disable thp for future mappings */
1219	thp_split_mm(mm);
1220	up_write(&mm->mmap_sem);
1221	return 0;
1222}
1223EXPORT_SYMBOL_GPL(s390_enable_sie);
1224
1225/*
1226 * Enable storage key handling from now on and initialize the storage
1227 * keys with the default key.
1228 */
1229static int __s390_enable_skey(pte_t *pte, unsigned long addr,
1230			      unsigned long next, struct mm_walk *walk)
1231{
1232	unsigned long ptev;
1233	pgste_t pgste;
1234
1235	pgste = pgste_get_lock(pte);
1236	/*
1237	 * Remove all zero page mappings,
1238	 * after establishing a policy to forbid zero page mappings
1239	 * following faults for that page will get fresh anonymous pages
1240	 */
1241	if (is_zero_pfn(pte_pfn(*pte))) {
1242		ptep_flush_direct(walk->mm, addr, pte);
1243		pte_val(*pte) = _PAGE_INVALID;
1244	}
1245	/* Clear storage key */
1246	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT |
1247			      PGSTE_GR_BIT | PGSTE_GC_BIT);
1248	ptev = pte_val(*pte);
1249	if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
1250		page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 1);
1251	pgste_set_unlock(pte, pgste);
1252	return 0;
1253}
1254
1255int s390_enable_skey(void)
1256{
1257	struct mm_walk walk = { .pte_entry = __s390_enable_skey };
1258	struct mm_struct *mm = current->mm;
1259	struct vm_area_struct *vma;
1260	int rc = 0;
1261
1262	down_write(&mm->mmap_sem);
1263	if (mm_use_skey(mm))
1264		goto out_up;
1265
1266	mm->context.use_skey = 1;
1267	for (vma = mm->mmap; vma; vma = vma->vm_next) {
1268		if (ksm_madvise(vma, vma->vm_start, vma->vm_end,
1269				MADV_UNMERGEABLE, &vma->vm_flags)) {
1270			mm->context.use_skey = 0;
1271			rc = -ENOMEM;
1272			goto out_up;
1273		}
1274	}
1275	mm->def_flags &= ~VM_MERGEABLE;
1276
1277	walk.mm = mm;
1278	walk_page_range(0, TASK_SIZE, &walk);
1279
1280out_up:
1281	up_write(&mm->mmap_sem);
1282	return rc;
1283}
1284EXPORT_SYMBOL_GPL(s390_enable_skey);
1285
1286/*
1287 * Reset CMMA state, make all pages stable again.
1288 */
1289static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
1290			     unsigned long next, struct mm_walk *walk)
1291{
1292	pgste_t pgste;
1293
1294	pgste = pgste_get_lock(pte);
1295	pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
1296	pgste_set_unlock(pte, pgste);
1297	return 0;
1298}
1299
1300void s390_reset_cmma(struct mm_struct *mm)
1301{
1302	struct mm_walk walk = { .pte_entry = __s390_reset_cmma };
1303
1304	down_write(&mm->mmap_sem);
1305	walk.mm = mm;
1306	walk_page_range(0, TASK_SIZE, &walk);
1307	up_write(&mm->mmap_sem);
1308}
1309EXPORT_SYMBOL_GPL(s390_reset_cmma);
1310
1311/*
1312 * Test and reset if a guest page is dirty
1313 */
1314bool gmap_test_and_clear_dirty(unsigned long address, struct gmap *gmap)
1315{
1316	pte_t *pte;
1317	spinlock_t *ptl;
1318	bool dirty = false;
1319
1320	pte = get_locked_pte(gmap->mm, address, &ptl);
1321	if (unlikely(!pte))
1322		return false;
1323
1324	if (ptep_test_and_clear_user_dirty(gmap->mm, address, pte))
1325		dirty = true;
1326
1327	spin_unlock(ptl);
1328	return dirty;
1329}
1330EXPORT_SYMBOL_GPL(gmap_test_and_clear_dirty);
1331
1332#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1333int pmdp_clear_flush_young(struct vm_area_struct *vma, unsigned long address,
1334			   pmd_t *pmdp)
1335{
1336	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
1337	/* No need to flush TLB
1338	 * On s390 reference bits are in storage key and never in TLB */
1339	return pmdp_test_and_clear_young(vma, address, pmdp);
1340}
1341
1342int pmdp_set_access_flags(struct vm_area_struct *vma,
1343			  unsigned long address, pmd_t *pmdp,
1344			  pmd_t entry, int dirty)
1345{
1346	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
1347
1348	entry = pmd_mkyoung(entry);
1349	if (dirty)
1350		entry = pmd_mkdirty(entry);
1351	if (pmd_same(*pmdp, entry))
1352		return 0;
1353	pmdp_invalidate(vma, address, pmdp);
1354	set_pmd_at(vma->vm_mm, address, pmdp, entry);
1355	return 1;
1356}
1357
1358static void pmdp_splitting_flush_sync(void *arg)
1359{
1360	/* Simply deliver the interrupt */
1361}
1362
1363void pmdp_splitting_flush(struct vm_area_struct *vma, unsigned long address,
1364			  pmd_t *pmdp)
1365{
1366	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
1367	if (!test_and_set_bit(_SEGMENT_ENTRY_SPLIT_BIT,
1368			      (unsigned long *) pmdp)) {
1369		/* need to serialize against gup-fast (IRQ disabled) */
1370		smp_call_function(pmdp_splitting_flush_sync, NULL, 1);
1371	}
1372}
1373
1374void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
1375				pgtable_t pgtable)
1376{
1377	struct list_head *lh = (struct list_head *) pgtable;
1378
1379	assert_spin_locked(pmd_lockptr(mm, pmdp));
1380
1381	/* FIFO */
1382	if (!pmd_huge_pte(mm, pmdp))
1383		INIT_LIST_HEAD(lh);
1384	else
1385		list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
1386	pmd_huge_pte(mm, pmdp) = pgtable;
1387}
1388
1389pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
1390{
1391	struct list_head *lh;
1392	pgtable_t pgtable;
1393	pte_t *ptep;
1394
1395	assert_spin_locked(pmd_lockptr(mm, pmdp));
1396
1397	/* FIFO */
1398	pgtable = pmd_huge_pte(mm, pmdp);
1399	lh = (struct list_head *) pgtable;
1400	if (list_empty(lh))
1401		pmd_huge_pte(mm, pmdp) = NULL;
1402	else {
1403		pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
1404		list_del(lh);
1405	}
1406	ptep = (pte_t *) pgtable;
1407	pte_val(*ptep) = _PAGE_INVALID;
1408	ptep++;
1409	pte_val(*ptep) = _PAGE_INVALID;
1410	return pgtable;
1411}
1412#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1413