Loading...
1/*
2 * Copyright IBM Corp. 2007,2009
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
36unsigned long VMALLOC_START = VMALLOC_END - VMALLOC_SIZE;
37EXPORT_SYMBOL(VMALLOC_START);
38
39static int __init parse_vmalloc(char *arg)
40{
41 if (!arg)
42 return -EINVAL;
43 VMALLOC_START = (VMALLOC_END - memparse(arg, &arg)) & PAGE_MASK;
44 return 0;
45}
46early_param("vmalloc", parse_vmalloc);
47
48unsigned long *crst_table_alloc(struct mm_struct *mm)
49{
50 struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
51
52 if (!page)
53 return NULL;
54 return (unsigned long *) page_to_phys(page);
55}
56
57void crst_table_free(struct mm_struct *mm, unsigned long *table)
58{
59 free_pages((unsigned long) table, ALLOC_ORDER);
60}
61
62#ifdef CONFIG_64BIT
63int crst_table_upgrade(struct mm_struct *mm, unsigned long limit)
64{
65 unsigned long *table, *pgd;
66 unsigned long entry;
67
68 BUG_ON(limit > (1UL << 53));
69repeat:
70 table = crst_table_alloc(mm);
71 if (!table)
72 return -ENOMEM;
73 spin_lock_bh(&mm->page_table_lock);
74 if (mm->context.asce_limit < limit) {
75 pgd = (unsigned long *) mm->pgd;
76 if (mm->context.asce_limit <= (1UL << 31)) {
77 entry = _REGION3_ENTRY_EMPTY;
78 mm->context.asce_limit = 1UL << 42;
79 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
80 _ASCE_USER_BITS |
81 _ASCE_TYPE_REGION3;
82 } else {
83 entry = _REGION2_ENTRY_EMPTY;
84 mm->context.asce_limit = 1UL << 53;
85 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
86 _ASCE_USER_BITS |
87 _ASCE_TYPE_REGION2;
88 }
89 crst_table_init(table, entry);
90 pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd);
91 mm->pgd = (pgd_t *) table;
92 mm->task_size = mm->context.asce_limit;
93 table = NULL;
94 }
95 spin_unlock_bh(&mm->page_table_lock);
96 if (table)
97 crst_table_free(mm, table);
98 if (mm->context.asce_limit < limit)
99 goto repeat;
100 update_mm(mm, current);
101 return 0;
102}
103
104void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
105{
106 pgd_t *pgd;
107
108 if (mm->context.asce_limit <= limit)
109 return;
110 __tlb_flush_mm(mm);
111 while (mm->context.asce_limit > limit) {
112 pgd = mm->pgd;
113 switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) {
114 case _REGION_ENTRY_TYPE_R2:
115 mm->context.asce_limit = 1UL << 42;
116 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
117 _ASCE_USER_BITS |
118 _ASCE_TYPE_REGION3;
119 break;
120 case _REGION_ENTRY_TYPE_R3:
121 mm->context.asce_limit = 1UL << 31;
122 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
123 _ASCE_USER_BITS |
124 _ASCE_TYPE_SEGMENT;
125 break;
126 default:
127 BUG();
128 }
129 mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
130 mm->task_size = mm->context.asce_limit;
131 crst_table_free(mm, (unsigned long *) pgd);
132 }
133 update_mm(mm, current);
134}
135#endif
136
137#ifdef CONFIG_PGSTE
138
139/**
140 * gmap_alloc - allocate a guest address space
141 * @mm: pointer to the parent mm_struct
142 *
143 * Returns a guest address space structure.
144 */
145struct gmap *gmap_alloc(struct mm_struct *mm)
146{
147 struct gmap *gmap;
148 struct page *page;
149 unsigned long *table;
150
151 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
152 if (!gmap)
153 goto out;
154 INIT_LIST_HEAD(&gmap->crst_list);
155 gmap->mm = mm;
156 page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
157 if (!page)
158 goto out_free;
159 list_add(&page->lru, &gmap->crst_list);
160 table = (unsigned long *) page_to_phys(page);
161 crst_table_init(table, _REGION1_ENTRY_EMPTY);
162 gmap->table = table;
163 gmap->asce = _ASCE_TYPE_REGION1 | _ASCE_TABLE_LENGTH |
164 _ASCE_USER_BITS | __pa(table);
165 list_add(&gmap->list, &mm->context.gmap_list);
166 return gmap;
167
168out_free:
169 kfree(gmap);
170out:
171 return NULL;
172}
173EXPORT_SYMBOL_GPL(gmap_alloc);
174
175static int gmap_unlink_segment(struct gmap *gmap, unsigned long *table)
176{
177 struct gmap_pgtable *mp;
178 struct gmap_rmap *rmap;
179 struct page *page;
180
181 if (*table & _SEGMENT_ENTRY_INV)
182 return 0;
183 page = pfn_to_page(*table >> PAGE_SHIFT);
184 mp = (struct gmap_pgtable *) page->index;
185 list_for_each_entry(rmap, &mp->mapper, list) {
186 if (rmap->entry != table)
187 continue;
188 list_del(&rmap->list);
189 kfree(rmap);
190 break;
191 }
192 *table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
193 return 1;
194}
195
196static void gmap_flush_tlb(struct gmap *gmap)
197{
198 if (MACHINE_HAS_IDTE)
199 __tlb_flush_idte((unsigned long) gmap->table |
200 _ASCE_TYPE_REGION1);
201 else
202 __tlb_flush_global();
203}
204
205/**
206 * gmap_free - free a guest address space
207 * @gmap: pointer to the guest address space structure
208 */
209void gmap_free(struct gmap *gmap)
210{
211 struct page *page, *next;
212 unsigned long *table;
213 int i;
214
215
216 /* Flush tlb. */
217 if (MACHINE_HAS_IDTE)
218 __tlb_flush_idte((unsigned long) gmap->table |
219 _ASCE_TYPE_REGION1);
220 else
221 __tlb_flush_global();
222
223 /* Free all segment & region tables. */
224 down_read(&gmap->mm->mmap_sem);
225 list_for_each_entry_safe(page, next, &gmap->crst_list, lru) {
226 table = (unsigned long *) page_to_phys(page);
227 if ((*table & _REGION_ENTRY_TYPE_MASK) == 0)
228 /* Remove gmap rmap structures for segment table. */
229 for (i = 0; i < PTRS_PER_PMD; i++, table++)
230 gmap_unlink_segment(gmap, table);
231 __free_pages(page, ALLOC_ORDER);
232 }
233 up_read(&gmap->mm->mmap_sem);
234 list_del(&gmap->list);
235 kfree(gmap);
236}
237EXPORT_SYMBOL_GPL(gmap_free);
238
239/**
240 * gmap_enable - switch primary space to the guest address space
241 * @gmap: pointer to the guest address space structure
242 */
243void gmap_enable(struct gmap *gmap)
244{
245 S390_lowcore.gmap = (unsigned long) gmap;
246}
247EXPORT_SYMBOL_GPL(gmap_enable);
248
249/**
250 * gmap_disable - switch back to the standard primary address space
251 * @gmap: pointer to the guest address space structure
252 */
253void gmap_disable(struct gmap *gmap)
254{
255 S390_lowcore.gmap = 0UL;
256}
257EXPORT_SYMBOL_GPL(gmap_disable);
258
259static int gmap_alloc_table(struct gmap *gmap,
260 unsigned long *table, unsigned long init)
261{
262 struct page *page;
263 unsigned long *new;
264
265 page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
266 if (!page)
267 return -ENOMEM;
268 new = (unsigned long *) page_to_phys(page);
269 crst_table_init(new, init);
270 down_read(&gmap->mm->mmap_sem);
271 if (*table & _REGION_ENTRY_INV) {
272 list_add(&page->lru, &gmap->crst_list);
273 *table = (unsigned long) new | _REGION_ENTRY_LENGTH |
274 (*table & _REGION_ENTRY_TYPE_MASK);
275 } else
276 __free_pages(page, ALLOC_ORDER);
277 up_read(&gmap->mm->mmap_sem);
278 return 0;
279}
280
281/**
282 * gmap_unmap_segment - unmap segment from the guest address space
283 * @gmap: pointer to the guest address space structure
284 * @addr: address in the guest address space
285 * @len: length of the memory area to unmap
286 *
287 * Returns 0 if the unmap succeded, -EINVAL if not.
288 */
289int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
290{
291 unsigned long *table;
292 unsigned long off;
293 int flush;
294
295 if ((to | len) & (PMD_SIZE - 1))
296 return -EINVAL;
297 if (len == 0 || to + len < to)
298 return -EINVAL;
299
300 flush = 0;
301 down_read(&gmap->mm->mmap_sem);
302 for (off = 0; off < len; off += PMD_SIZE) {
303 /* Walk the guest addr space page table */
304 table = gmap->table + (((to + off) >> 53) & 0x7ff);
305 if (*table & _REGION_ENTRY_INV)
306 goto out;
307 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
308 table = table + (((to + off) >> 42) & 0x7ff);
309 if (*table & _REGION_ENTRY_INV)
310 goto out;
311 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
312 table = table + (((to + off) >> 31) & 0x7ff);
313 if (*table & _REGION_ENTRY_INV)
314 goto out;
315 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
316 table = table + (((to + off) >> 20) & 0x7ff);
317
318 /* Clear segment table entry in guest address space. */
319 flush |= gmap_unlink_segment(gmap, table);
320 *table = _SEGMENT_ENTRY_INV;
321 }
322out:
323 up_read(&gmap->mm->mmap_sem);
324 if (flush)
325 gmap_flush_tlb(gmap);
326 return 0;
327}
328EXPORT_SYMBOL_GPL(gmap_unmap_segment);
329
330/**
331 * gmap_mmap_segment - map a segment to the guest address space
332 * @gmap: pointer to the guest address space structure
333 * @from: source address in the parent address space
334 * @to: target address in the guest address space
335 *
336 * Returns 0 if the mmap succeded, -EINVAL or -ENOMEM if not.
337 */
338int gmap_map_segment(struct gmap *gmap, unsigned long from,
339 unsigned long to, unsigned long len)
340{
341 unsigned long *table;
342 unsigned long off;
343 int flush;
344
345 if ((from | to | len) & (PMD_SIZE - 1))
346 return -EINVAL;
347 if (len == 0 || from + len > PGDIR_SIZE ||
348 from + len < from || to + len < to)
349 return -EINVAL;
350
351 flush = 0;
352 down_read(&gmap->mm->mmap_sem);
353 for (off = 0; off < len; off += PMD_SIZE) {
354 /* Walk the gmap address space page table */
355 table = gmap->table + (((to + off) >> 53) & 0x7ff);
356 if ((*table & _REGION_ENTRY_INV) &&
357 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY))
358 goto out_unmap;
359 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
360 table = table + (((to + off) >> 42) & 0x7ff);
361 if ((*table & _REGION_ENTRY_INV) &&
362 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY))
363 goto out_unmap;
364 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
365 table = table + (((to + off) >> 31) & 0x7ff);
366 if ((*table & _REGION_ENTRY_INV) &&
367 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY))
368 goto out_unmap;
369 table = (unsigned long *) (*table & _REGION_ENTRY_ORIGIN);
370 table = table + (((to + off) >> 20) & 0x7ff);
371
372 /* Store 'from' address in an invalid segment table entry. */
373 flush |= gmap_unlink_segment(gmap, table);
374 *table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | (from + off);
375 }
376 up_read(&gmap->mm->mmap_sem);
377 if (flush)
378 gmap_flush_tlb(gmap);
379 return 0;
380
381out_unmap:
382 up_read(&gmap->mm->mmap_sem);
383 gmap_unmap_segment(gmap, to, len);
384 return -ENOMEM;
385}
386EXPORT_SYMBOL_GPL(gmap_map_segment);
387
388unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
389{
390 unsigned long *table, vmaddr, segment;
391 struct mm_struct *mm;
392 struct gmap_pgtable *mp;
393 struct gmap_rmap *rmap;
394 struct vm_area_struct *vma;
395 struct page *page;
396 pgd_t *pgd;
397 pud_t *pud;
398 pmd_t *pmd;
399
400 current->thread.gmap_addr = address;
401 mm = gmap->mm;
402 /* Walk the gmap address space page table */
403 table = gmap->table + ((address >> 53) & 0x7ff);
404 if (unlikely(*table & _REGION_ENTRY_INV))
405 return -EFAULT;
406 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
407 table = table + ((address >> 42) & 0x7ff);
408 if (unlikely(*table & _REGION_ENTRY_INV))
409 return -EFAULT;
410 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
411 table = table + ((address >> 31) & 0x7ff);
412 if (unlikely(*table & _REGION_ENTRY_INV))
413 return -EFAULT;
414 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
415 table = table + ((address >> 20) & 0x7ff);
416
417 /* Convert the gmap address to an mm address. */
418 segment = *table;
419 if (likely(!(segment & _SEGMENT_ENTRY_INV))) {
420 page = pfn_to_page(segment >> PAGE_SHIFT);
421 mp = (struct gmap_pgtable *) page->index;
422 return mp->vmaddr | (address & ~PMD_MASK);
423 } else if (segment & _SEGMENT_ENTRY_RO) {
424 vmaddr = segment & _SEGMENT_ENTRY_ORIGIN;
425 vma = find_vma(mm, vmaddr);
426 if (!vma || vma->vm_start > vmaddr)
427 return -EFAULT;
428
429 /* Walk the parent mm page table */
430 pgd = pgd_offset(mm, vmaddr);
431 pud = pud_alloc(mm, pgd, vmaddr);
432 if (!pud)
433 return -ENOMEM;
434 pmd = pmd_alloc(mm, pud, vmaddr);
435 if (!pmd)
436 return -ENOMEM;
437 if (!pmd_present(*pmd) &&
438 __pte_alloc(mm, vma, pmd, vmaddr))
439 return -ENOMEM;
440 /* pmd now points to a valid segment table entry. */
441 rmap = kmalloc(sizeof(*rmap), GFP_KERNEL|__GFP_REPEAT);
442 if (!rmap)
443 return -ENOMEM;
444 /* Link gmap segment table entry location to page table. */
445 page = pmd_page(*pmd);
446 mp = (struct gmap_pgtable *) page->index;
447 rmap->entry = table;
448 list_add(&rmap->list, &mp->mapper);
449 /* Set gmap segment table entry to page table. */
450 *table = pmd_val(*pmd) & PAGE_MASK;
451 return vmaddr | (address & ~PMD_MASK);
452 }
453 return -EFAULT;
454
455}
456EXPORT_SYMBOL_GPL(gmap_fault);
457
458void gmap_unmap_notifier(struct mm_struct *mm, unsigned long *table)
459{
460 struct gmap_rmap *rmap, *next;
461 struct gmap_pgtable *mp;
462 struct page *page;
463 int flush;
464
465 flush = 0;
466 spin_lock(&mm->page_table_lock);
467 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
468 mp = (struct gmap_pgtable *) page->index;
469 list_for_each_entry_safe(rmap, next, &mp->mapper, list) {
470 *rmap->entry =
471 _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
472 list_del(&rmap->list);
473 kfree(rmap);
474 flush = 1;
475 }
476 spin_unlock(&mm->page_table_lock);
477 if (flush)
478 __tlb_flush_global();
479}
480
481static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
482 unsigned long vmaddr)
483{
484 struct page *page;
485 unsigned long *table;
486 struct gmap_pgtable *mp;
487
488 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
489 if (!page)
490 return NULL;
491 mp = kmalloc(sizeof(*mp), GFP_KERNEL|__GFP_REPEAT);
492 if (!mp) {
493 __free_page(page);
494 return NULL;
495 }
496 pgtable_page_ctor(page);
497 mp->vmaddr = vmaddr & PMD_MASK;
498 INIT_LIST_HEAD(&mp->mapper);
499 page->index = (unsigned long) mp;
500 atomic_set(&page->_mapcount, 3);
501 table = (unsigned long *) page_to_phys(page);
502 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/2);
503 clear_table(table + PTRS_PER_PTE, 0, PAGE_SIZE/2);
504 return table;
505}
506
507static inline void page_table_free_pgste(unsigned long *table)
508{
509 struct page *page;
510 struct gmap_pgtable *mp;
511
512 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
513 mp = (struct gmap_pgtable *) page->index;
514 BUG_ON(!list_empty(&mp->mapper));
515 pgtable_page_ctor(page);
516 atomic_set(&page->_mapcount, -1);
517 kfree(mp);
518 __free_page(page);
519}
520
521#else /* CONFIG_PGSTE */
522
523static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
524 unsigned long vmaddr)
525{
526 return NULL;
527}
528
529static inline void page_table_free_pgste(unsigned long *table)
530{
531}
532
533static inline void gmap_unmap_notifier(struct mm_struct *mm,
534 unsigned long *table)
535{
536}
537
538#endif /* CONFIG_PGSTE */
539
540static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
541{
542 unsigned int old, new;
543
544 do {
545 old = atomic_read(v);
546 new = old ^ bits;
547 } while (atomic_cmpxchg(v, old, new) != old);
548 return new;
549}
550
551/*
552 * page table entry allocation/free routines.
553 */
554unsigned long *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr)
555{
556 struct page *page;
557 unsigned long *table;
558 unsigned int mask, bit;
559
560 if (mm_has_pgste(mm))
561 return page_table_alloc_pgste(mm, vmaddr);
562 /* Allocate fragments of a 4K page as 1K/2K page table */
563 spin_lock_bh(&mm->context.list_lock);
564 mask = FRAG_MASK;
565 if (!list_empty(&mm->context.pgtable_list)) {
566 page = list_first_entry(&mm->context.pgtable_list,
567 struct page, lru);
568 table = (unsigned long *) page_to_phys(page);
569 mask = atomic_read(&page->_mapcount);
570 mask = mask | (mask >> 4);
571 }
572 if ((mask & FRAG_MASK) == FRAG_MASK) {
573 spin_unlock_bh(&mm->context.list_lock);
574 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
575 if (!page)
576 return NULL;
577 pgtable_page_ctor(page);
578 atomic_set(&page->_mapcount, 1);
579 table = (unsigned long *) page_to_phys(page);
580 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE);
581 spin_lock_bh(&mm->context.list_lock);
582 list_add(&page->lru, &mm->context.pgtable_list);
583 } else {
584 for (bit = 1; mask & bit; bit <<= 1)
585 table += PTRS_PER_PTE;
586 mask = atomic_xor_bits(&page->_mapcount, bit);
587 if ((mask & FRAG_MASK) == FRAG_MASK)
588 list_del(&page->lru);
589 }
590 spin_unlock_bh(&mm->context.list_lock);
591 return table;
592}
593
594void page_table_free(struct mm_struct *mm, unsigned long *table)
595{
596 struct page *page;
597 unsigned int bit, mask;
598
599 if (mm_has_pgste(mm)) {
600 gmap_unmap_notifier(mm, table);
601 return page_table_free_pgste(table);
602 }
603 /* Free 1K/2K page table fragment of a 4K page */
604 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
605 bit = 1 << ((__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t)));
606 spin_lock_bh(&mm->context.list_lock);
607 if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
608 list_del(&page->lru);
609 mask = atomic_xor_bits(&page->_mapcount, bit);
610 if (mask & FRAG_MASK)
611 list_add(&page->lru, &mm->context.pgtable_list);
612 spin_unlock_bh(&mm->context.list_lock);
613 if (mask == 0) {
614 pgtable_page_dtor(page);
615 atomic_set(&page->_mapcount, -1);
616 __free_page(page);
617 }
618}
619
620#ifdef CONFIG_HAVE_RCU_TABLE_FREE
621
622static void __page_table_free_rcu(void *table, unsigned bit)
623{
624 struct page *page;
625
626 if (bit == FRAG_MASK)
627 return page_table_free_pgste(table);
628 /* Free 1K/2K page table fragment of a 4K page */
629 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
630 if (atomic_xor_bits(&page->_mapcount, bit) == 0) {
631 pgtable_page_dtor(page);
632 atomic_set(&page->_mapcount, -1);
633 __free_page(page);
634 }
635}
636
637void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table)
638{
639 struct mm_struct *mm;
640 struct page *page;
641 unsigned int bit, mask;
642
643 mm = tlb->mm;
644 if (mm_has_pgste(mm)) {
645 gmap_unmap_notifier(mm, table);
646 table = (unsigned long *) (__pa(table) | FRAG_MASK);
647 tlb_remove_table(tlb, table);
648 return;
649 }
650 bit = 1 << ((__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t)));
651 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
652 spin_lock_bh(&mm->context.list_lock);
653 if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
654 list_del(&page->lru);
655 mask = atomic_xor_bits(&page->_mapcount, bit | (bit << 4));
656 if (mask & FRAG_MASK)
657 list_add_tail(&page->lru, &mm->context.pgtable_list);
658 spin_unlock_bh(&mm->context.list_lock);
659 table = (unsigned long *) (__pa(table) | (bit << 4));
660 tlb_remove_table(tlb, table);
661}
662
663void __tlb_remove_table(void *_table)
664{
665 void *table = (void *)((unsigned long) _table & PAGE_MASK);
666 unsigned type = (unsigned long) _table & ~PAGE_MASK;
667
668 if (type)
669 __page_table_free_rcu(table, type);
670 else
671 free_pages((unsigned long) table, ALLOC_ORDER);
672}
673
674#endif
675
676/*
677 * switch on pgstes for its userspace process (for kvm)
678 */
679int s390_enable_sie(void)
680{
681 struct task_struct *tsk = current;
682 struct mm_struct *mm, *old_mm;
683
684 /* Do we have switched amode? If no, we cannot do sie */
685 if (user_mode == HOME_SPACE_MODE)
686 return -EINVAL;
687
688 /* Do we have pgstes? if yes, we are done */
689 if (mm_has_pgste(tsk->mm))
690 return 0;
691
692 /* lets check if we are allowed to replace the mm */
693 task_lock(tsk);
694 if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
695#ifdef CONFIG_AIO
696 !hlist_empty(&tsk->mm->ioctx_list) ||
697#endif
698 tsk->mm != tsk->active_mm) {
699 task_unlock(tsk);
700 return -EINVAL;
701 }
702 task_unlock(tsk);
703
704 /* we copy the mm and let dup_mm create the page tables with_pgstes */
705 tsk->mm->context.alloc_pgste = 1;
706 mm = dup_mm(tsk);
707 tsk->mm->context.alloc_pgste = 0;
708 if (!mm)
709 return -ENOMEM;
710
711 /* Now lets check again if something happened */
712 task_lock(tsk);
713 if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
714#ifdef CONFIG_AIO
715 !hlist_empty(&tsk->mm->ioctx_list) ||
716#endif
717 tsk->mm != tsk->active_mm) {
718 mmput(mm);
719 task_unlock(tsk);
720 return -EINVAL;
721 }
722
723 /* ok, we are alone. No ptrace, no threads, etc. */
724 old_mm = tsk->mm;
725 tsk->mm = tsk->active_mm = mm;
726 preempt_disable();
727 update_mm(mm, tsk);
728 atomic_inc(&mm->context.attach_count);
729 atomic_dec(&old_mm->context.attach_count);
730 cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
731 preempt_enable();
732 task_unlock(tsk);
733 mmput(old_mm);
734 return 0;
735}
736EXPORT_SYMBOL_GPL(s390_enable_sie);
737
738#if defined(CONFIG_DEBUG_PAGEALLOC) && defined(CONFIG_HIBERNATION)
739bool kernel_page_present(struct page *page)
740{
741 unsigned long addr;
742 int cc;
743
744 addr = page_to_phys(page);
745 asm volatile(
746 " lra %1,0(%1)\n"
747 " ipm %0\n"
748 " srl %0,28"
749 : "=d" (cc), "+a" (addr) : : "cc");
750 return cc == 0;
751}
752#endif /* CONFIG_HIBERNATION && CONFIG_DEBUG_PAGEALLOC */
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/spinlock.h>
14#include <linux/rcupdate.h>
15#include <linux/slab.h>
16#include <linux/swapops.h>
17#include <linux/sysctl.h>
18#include <linux/ksm.h>
19#include <linux/mman.h>
20
21#include <asm/pgtable.h>
22#include <asm/pgalloc.h>
23#include <asm/tlb.h>
24#include <asm/tlbflush.h>
25#include <asm/mmu_context.h>
26
27static inline pte_t ptep_flush_direct(struct mm_struct *mm,
28 unsigned long addr, pte_t *ptep)
29{
30 int active, count;
31 pte_t old;
32
33 old = *ptep;
34 if (unlikely(pte_val(old) & _PAGE_INVALID))
35 return old;
36 active = (mm == current->active_mm) ? 1 : 0;
37 count = atomic_add_return(0x10000, &mm->context.attach_count);
38 if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
39 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
40 __ptep_ipte_local(addr, ptep);
41 else
42 __ptep_ipte(addr, ptep);
43 atomic_sub(0x10000, &mm->context.attach_count);
44 return old;
45}
46
47static inline pte_t ptep_flush_lazy(struct mm_struct *mm,
48 unsigned long addr, pte_t *ptep)
49{
50 int active, count;
51 pte_t old;
52
53 old = *ptep;
54 if (unlikely(pte_val(old) & _PAGE_INVALID))
55 return old;
56 active = (mm == current->active_mm) ? 1 : 0;
57 count = atomic_add_return(0x10000, &mm->context.attach_count);
58 if ((count & 0xffff) <= active) {
59 pte_val(*ptep) |= _PAGE_INVALID;
60 mm->context.flush_mm = 1;
61 } else
62 __ptep_ipte(addr, ptep);
63 atomic_sub(0x10000, &mm->context.attach_count);
64 return old;
65}
66
67static inline pgste_t pgste_get_lock(pte_t *ptep)
68{
69 unsigned long new = 0;
70#ifdef CONFIG_PGSTE
71 unsigned long old;
72
73 preempt_disable();
74 asm(
75 " lg %0,%2\n"
76 "0: lgr %1,%0\n"
77 " nihh %0,0xff7f\n" /* clear PCL bit in old */
78 " oihh %1,0x0080\n" /* set PCL bit in new */
79 " csg %0,%1,%2\n"
80 " jl 0b\n"
81 : "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
82 : "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
83#endif
84 return __pgste(new);
85}
86
87static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
88{
89#ifdef CONFIG_PGSTE
90 asm(
91 " nihh %1,0xff7f\n" /* clear PCL bit */
92 " stg %1,%0\n"
93 : "=Q" (ptep[PTRS_PER_PTE])
94 : "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
95 : "cc", "memory");
96 preempt_enable();
97#endif
98}
99
100static inline pgste_t pgste_get(pte_t *ptep)
101{
102 unsigned long pgste = 0;
103#ifdef CONFIG_PGSTE
104 pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
105#endif
106 return __pgste(pgste);
107}
108
109static inline void pgste_set(pte_t *ptep, pgste_t pgste)
110{
111#ifdef CONFIG_PGSTE
112 *(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
113#endif
114}
115
116static inline pgste_t pgste_update_all(pte_t pte, pgste_t pgste,
117 struct mm_struct *mm)
118{
119#ifdef CONFIG_PGSTE
120 unsigned long address, bits, skey;
121
122 if (!mm_use_skey(mm) || pte_val(pte) & _PAGE_INVALID)
123 return pgste;
124 address = pte_val(pte) & PAGE_MASK;
125 skey = (unsigned long) page_get_storage_key(address);
126 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
127 /* Transfer page changed & referenced bit to guest bits in pgste */
128 pgste_val(pgste) |= bits << 48; /* GR bit & GC bit */
129 /* Copy page access key and fetch protection bit to pgste */
130 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
131 pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
132#endif
133 return pgste;
134
135}
136
137static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
138 struct mm_struct *mm)
139{
140#ifdef CONFIG_PGSTE
141 unsigned long address;
142 unsigned long nkey;
143
144 if (!mm_use_skey(mm) || pte_val(entry) & _PAGE_INVALID)
145 return;
146 VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
147 address = pte_val(entry) & PAGE_MASK;
148 /*
149 * Set page access key and fetch protection bit from pgste.
150 * The guest C/R information is still in the PGSTE, set real
151 * key C/R to 0.
152 */
153 nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
154 nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
155 page_set_storage_key(address, nkey, 0);
156#endif
157}
158
159static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
160{
161#ifdef CONFIG_PGSTE
162 if ((pte_val(entry) & _PAGE_PRESENT) &&
163 (pte_val(entry) & _PAGE_WRITE) &&
164 !(pte_val(entry) & _PAGE_INVALID)) {
165 if (!MACHINE_HAS_ESOP) {
166 /*
167 * Without enhanced suppression-on-protection force
168 * the dirty bit on for all writable ptes.
169 */
170 pte_val(entry) |= _PAGE_DIRTY;
171 pte_val(entry) &= ~_PAGE_PROTECT;
172 }
173 if (!(pte_val(entry) & _PAGE_PROTECT))
174 /* This pte allows write access, set user-dirty */
175 pgste_val(pgste) |= PGSTE_UC_BIT;
176 }
177#endif
178 *ptep = entry;
179 return pgste;
180}
181
182static inline pgste_t pgste_ipte_notify(struct mm_struct *mm,
183 unsigned long addr,
184 pte_t *ptep, pgste_t pgste)
185{
186#ifdef CONFIG_PGSTE
187 if (pgste_val(pgste) & PGSTE_IN_BIT) {
188 pgste_val(pgste) &= ~PGSTE_IN_BIT;
189 ptep_notify(mm, addr, ptep);
190 }
191#endif
192 return pgste;
193}
194
195static inline pgste_t ptep_xchg_start(struct mm_struct *mm,
196 unsigned long addr, pte_t *ptep)
197{
198 pgste_t pgste = __pgste(0);
199
200 if (mm_has_pgste(mm)) {
201 pgste = pgste_get_lock(ptep);
202 pgste = pgste_ipte_notify(mm, addr, ptep, pgste);
203 }
204 return pgste;
205}
206
207static inline void ptep_xchg_commit(struct mm_struct *mm,
208 unsigned long addr, pte_t *ptep,
209 pgste_t pgste, pte_t old, pte_t new)
210{
211 if (mm_has_pgste(mm)) {
212 if (pte_val(old) & _PAGE_INVALID)
213 pgste_set_key(ptep, pgste, new, mm);
214 if (pte_val(new) & _PAGE_INVALID) {
215 pgste = pgste_update_all(old, pgste, mm);
216 if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
217 _PGSTE_GPS_USAGE_UNUSED)
218 pte_val(old) |= _PAGE_UNUSED;
219 }
220 pgste = pgste_set_pte(ptep, pgste, new);
221 pgste_set_unlock(ptep, pgste);
222 } else {
223 *ptep = new;
224 }
225}
226
227pte_t ptep_xchg_direct(struct mm_struct *mm, unsigned long addr,
228 pte_t *ptep, pte_t new)
229{
230 pgste_t pgste;
231 pte_t old;
232
233 pgste = ptep_xchg_start(mm, addr, ptep);
234 old = ptep_flush_direct(mm, addr, ptep);
235 ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
236 return old;
237}
238EXPORT_SYMBOL(ptep_xchg_direct);
239
240pte_t ptep_xchg_lazy(struct mm_struct *mm, unsigned long addr,
241 pte_t *ptep, pte_t new)
242{
243 pgste_t pgste;
244 pte_t old;
245
246 pgste = ptep_xchg_start(mm, addr, ptep);
247 old = ptep_flush_lazy(mm, addr, ptep);
248 ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
249 return old;
250}
251EXPORT_SYMBOL(ptep_xchg_lazy);
252
253pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
254 pte_t *ptep)
255{
256 pgste_t pgste;
257 pte_t old;
258
259 pgste = ptep_xchg_start(mm, addr, ptep);
260 old = ptep_flush_lazy(mm, addr, ptep);
261 if (mm_has_pgste(mm)) {
262 pgste = pgste_update_all(old, pgste, mm);
263 pgste_set(ptep, pgste);
264 }
265 return old;
266}
267EXPORT_SYMBOL(ptep_modify_prot_start);
268
269void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
270 pte_t *ptep, pte_t pte)
271{
272 pgste_t pgste;
273
274 if (mm_has_pgste(mm)) {
275 pgste = pgste_get(ptep);
276 pgste_set_key(ptep, pgste, pte, mm);
277 pgste = pgste_set_pte(ptep, pgste, pte);
278 pgste_set_unlock(ptep, pgste);
279 } else {
280 *ptep = pte;
281 }
282}
283EXPORT_SYMBOL(ptep_modify_prot_commit);
284
285static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
286 unsigned long addr, pmd_t *pmdp)
287{
288 int active, count;
289 pmd_t old;
290
291 old = *pmdp;
292 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
293 return old;
294 if (!MACHINE_HAS_IDTE) {
295 __pmdp_csp(pmdp);
296 return old;
297 }
298 active = (mm == current->active_mm) ? 1 : 0;
299 count = atomic_add_return(0x10000, &mm->context.attach_count);
300 if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
301 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
302 __pmdp_idte_local(addr, pmdp);
303 else
304 __pmdp_idte(addr, pmdp);
305 atomic_sub(0x10000, &mm->context.attach_count);
306 return old;
307}
308
309static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
310 unsigned long addr, pmd_t *pmdp)
311{
312 int active, count;
313 pmd_t old;
314
315 old = *pmdp;
316 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
317 return old;
318 active = (mm == current->active_mm) ? 1 : 0;
319 count = atomic_add_return(0x10000, &mm->context.attach_count);
320 if ((count & 0xffff) <= active) {
321 pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
322 mm->context.flush_mm = 1;
323 } else if (MACHINE_HAS_IDTE)
324 __pmdp_idte(addr, pmdp);
325 else
326 __pmdp_csp(pmdp);
327 atomic_sub(0x10000, &mm->context.attach_count);
328 return old;
329}
330
331pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
332 pmd_t *pmdp, pmd_t new)
333{
334 pmd_t old;
335
336 old = pmdp_flush_direct(mm, addr, pmdp);
337 *pmdp = new;
338 return old;
339}
340EXPORT_SYMBOL(pmdp_xchg_direct);
341
342pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
343 pmd_t *pmdp, pmd_t new)
344{
345 pmd_t old;
346
347 old = pmdp_flush_lazy(mm, addr, pmdp);
348 *pmdp = new;
349 return old;
350}
351EXPORT_SYMBOL(pmdp_xchg_lazy);
352
353#ifdef CONFIG_TRANSPARENT_HUGEPAGE
354void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
355 pgtable_t pgtable)
356{
357 struct list_head *lh = (struct list_head *) pgtable;
358
359 assert_spin_locked(pmd_lockptr(mm, pmdp));
360
361 /* FIFO */
362 if (!pmd_huge_pte(mm, pmdp))
363 INIT_LIST_HEAD(lh);
364 else
365 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
366 pmd_huge_pte(mm, pmdp) = pgtable;
367}
368
369pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
370{
371 struct list_head *lh;
372 pgtable_t pgtable;
373 pte_t *ptep;
374
375 assert_spin_locked(pmd_lockptr(mm, pmdp));
376
377 /* FIFO */
378 pgtable = pmd_huge_pte(mm, pmdp);
379 lh = (struct list_head *) pgtable;
380 if (list_empty(lh))
381 pmd_huge_pte(mm, pmdp) = NULL;
382 else {
383 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
384 list_del(lh);
385 }
386 ptep = (pte_t *) pgtable;
387 pte_val(*ptep) = _PAGE_INVALID;
388 ptep++;
389 pte_val(*ptep) = _PAGE_INVALID;
390 return pgtable;
391}
392#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
393
394#ifdef CONFIG_PGSTE
395void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
396 pte_t *ptep, pte_t entry)
397{
398 pgste_t pgste;
399
400 /* the mm_has_pgste() check is done in set_pte_at() */
401 pgste = pgste_get_lock(ptep);
402 pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
403 pgste_set_key(ptep, pgste, entry, mm);
404 pgste = pgste_set_pte(ptep, pgste, entry);
405 pgste_set_unlock(ptep, pgste);
406}
407
408void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
409{
410 pgste_t pgste;
411
412 pgste = pgste_get_lock(ptep);
413 pgste_val(pgste) |= PGSTE_IN_BIT;
414 pgste_set_unlock(ptep, pgste);
415}
416
417static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
418{
419 if (!non_swap_entry(entry))
420 dec_mm_counter(mm, MM_SWAPENTS);
421 else if (is_migration_entry(entry)) {
422 struct page *page = migration_entry_to_page(entry);
423
424 dec_mm_counter(mm, mm_counter(page));
425 }
426 free_swap_and_cache(entry);
427}
428
429void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
430 pte_t *ptep, int reset)
431{
432 unsigned long pgstev;
433 pgste_t pgste;
434 pte_t pte;
435
436 /* Zap unused and logically-zero pages */
437 pgste = pgste_get_lock(ptep);
438 pgstev = pgste_val(pgste);
439 pte = *ptep;
440 if (pte_swap(pte) &&
441 ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
442 (pgstev & _PGSTE_GPS_ZERO))) {
443 ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
444 pte_clear(mm, addr, ptep);
445 }
446 if (reset)
447 pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
448 pgste_set_unlock(ptep, pgste);
449}
450
451void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
452{
453 unsigned long ptev;
454 pgste_t pgste;
455
456 /* Clear storage key */
457 pgste = pgste_get_lock(ptep);
458 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT |
459 PGSTE_GR_BIT | PGSTE_GC_BIT);
460 ptev = pte_val(*ptep);
461 if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
462 page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 1);
463 pgste_set_unlock(ptep, pgste);
464}
465
466/*
467 * Test and reset if a guest page is dirty
468 */
469bool test_and_clear_guest_dirty(struct mm_struct *mm, unsigned long addr)
470{
471 spinlock_t *ptl;
472 pgste_t pgste;
473 pte_t *ptep;
474 pte_t pte;
475 bool dirty;
476
477 ptep = get_locked_pte(mm, addr, &ptl);
478 if (unlikely(!ptep))
479 return false;
480
481 pgste = pgste_get_lock(ptep);
482 dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
483 pgste_val(pgste) &= ~PGSTE_UC_BIT;
484 pte = *ptep;
485 if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
486 pgste = pgste_ipte_notify(mm, addr, ptep, pgste);
487 __ptep_ipte(addr, ptep);
488 if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
489 pte_val(pte) |= _PAGE_PROTECT;
490 else
491 pte_val(pte) |= _PAGE_INVALID;
492 *ptep = pte;
493 }
494 pgste_set_unlock(ptep, pgste);
495
496 spin_unlock(ptl);
497 return dirty;
498}
499EXPORT_SYMBOL_GPL(test_and_clear_guest_dirty);
500
501int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
502 unsigned char key, bool nq)
503{
504 unsigned long keyul;
505 spinlock_t *ptl;
506 pgste_t old, new;
507 pte_t *ptep;
508
509 down_read(&mm->mmap_sem);
510 ptep = get_locked_pte(mm, addr, &ptl);
511 if (unlikely(!ptep)) {
512 up_read(&mm->mmap_sem);
513 return -EFAULT;
514 }
515
516 new = old = pgste_get_lock(ptep);
517 pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
518 PGSTE_ACC_BITS | PGSTE_FP_BIT);
519 keyul = (unsigned long) key;
520 pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
521 pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
522 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
523 unsigned long address, bits, skey;
524
525 address = pte_val(*ptep) & PAGE_MASK;
526 skey = (unsigned long) page_get_storage_key(address);
527 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
528 skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
529 /* Set storage key ACC and FP */
530 page_set_storage_key(address, skey, !nq);
531 /* Merge host changed & referenced into pgste */
532 pgste_val(new) |= bits << 52;
533 }
534 /* changing the guest storage key is considered a change of the page */
535 if ((pgste_val(new) ^ pgste_val(old)) &
536 (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
537 pgste_val(new) |= PGSTE_UC_BIT;
538
539 pgste_set_unlock(ptep, new);
540 pte_unmap_unlock(ptep, ptl);
541 up_read(&mm->mmap_sem);
542 return 0;
543}
544EXPORT_SYMBOL(set_guest_storage_key);
545
546unsigned char get_guest_storage_key(struct mm_struct *mm, unsigned long addr)
547{
548 unsigned char key;
549 spinlock_t *ptl;
550 pgste_t pgste;
551 pte_t *ptep;
552
553 down_read(&mm->mmap_sem);
554 ptep = get_locked_pte(mm, addr, &ptl);
555 if (unlikely(!ptep)) {
556 up_read(&mm->mmap_sem);
557 return -EFAULT;
558 }
559 pgste = pgste_get_lock(ptep);
560
561 if (pte_val(*ptep) & _PAGE_INVALID) {
562 key = (pgste_val(pgste) & PGSTE_ACC_BITS) >> 56;
563 key |= (pgste_val(pgste) & PGSTE_FP_BIT) >> 56;
564 key |= (pgste_val(pgste) & PGSTE_GR_BIT) >> 48;
565 key |= (pgste_val(pgste) & PGSTE_GC_BIT) >> 48;
566 } else {
567 key = page_get_storage_key(pte_val(*ptep) & PAGE_MASK);
568
569 /* Reflect guest's logical view, not physical */
570 if (pgste_val(pgste) & PGSTE_GR_BIT)
571 key |= _PAGE_REFERENCED;
572 if (pgste_val(pgste) & PGSTE_GC_BIT)
573 key |= _PAGE_CHANGED;
574 }
575
576 pgste_set_unlock(ptep, pgste);
577 pte_unmap_unlock(ptep, ptl);
578 up_read(&mm->mmap_sem);
579 return key;
580}
581EXPORT_SYMBOL(get_guest_storage_key);
582#endif